Errors resulting from the implementation of PMTs
Exclusion by design: An assessment of the effectiveness of the proxy means test poverty targeting mechanism
13
errors into the targeting process. Many of those households that may have been ‘correctly
targeted ’ in the first year are likely to be ‘inclusion errors’ in future years, as a result of
improved circumstances. However, anyone falling into poverty between surveys – perhaps
due to a crisis such as unemployment, disability, ill-health, death of a breadwinner or even the birth of a child
– is excluded from accessing social protection no matter how challenging their circumstances.
In Pakistan’s Benazir Income Support Programme BISP, for example, only 22 per cent of beneficiaries were living in poverty across each
of three survey years 2011, 2013 and 2014, 20 per cent were never under the poverty line while almost 60 per cent spent one or two of the three years in poverty, remaining on the
programme when above the poverty line Cheema et al, 2015. On the other hand, the vast majority of the population spending some time living in poverty over the same period were
unable to access BISP.
Figure 10 illustrates the challenges posed by changes in household composition, income, consumption and assets. It uses a panel dataset in Uganda to show who
– in red – would be accurately selected by a PMT targeting the poorest 20 per cent of households in
2011 in the left hand graph. The right hand graph indicates where those same households would be in 2013, if they were retargeted. Only 55 per cent of those accurately targeted in
2011 would still be accurately targeted in 2013: around 38 per cent would no longer be in the poorest 20 per cent in terms of expenditure, while 21 per cent would have PMT scores
above the cut-off. Many other households
– in blue – would be newly eligible for the programme by 2013, both in terms of consumption and their PMT score, but would not be
included, unless there were a new registration process.
Figure 10: Scattergraphs showing how households accurately targeted in 2011 would be targeted using the same PMT in 2013 Uganda
Source: Analysis undertaken by the authors using UNPS 201112 and UNPS 201314 datasets. See: Kidd and Bailey- Athias 2016.
However, despite the reality of significant changes in household incomes over short periods of time, it is very rare for PMT surveys to be repeated on a frequent basis.
Governments often perceive PMT surveys to be expensive: for example, the 2009 PMT survey in Pakistan and the 2011 survey in Indonesia both cost around US60 million, while
the 2015 survey in Indonesia cost US100 million.
10
Indeed, some PMT surveys are very expensive: for example, Kenya’s HSNP programme spent around US10 million to survey
only 380,000 households while achieving – as indicated earlier – not much better than
10
While these costs may seem high, they are, in reality, low when compared to the costs of implementing a more effective and robust targeting mechanism in a developing country.
14
Exclusion by design: An assessment of the effectiveness of the proxy means test poverty targeting mechanism
random selection Fitzgibbon, 2014. As a result of the perceived high costs of PMT surveys, governments are reluctant to repeat them. Pakistan, for example, has not
undertaken a PMT survey since 2009; in Indonesia, there was a four-year gap between the surveys of 2011 and 2015; while, in some areas of Mexico, registration for the
Oportunidades
programme had not been repeated for more than 10 years Zoletto, 2011. Furthermore, both the infrequency of information and the static nature of the PMT
means that programmes using a PMT cannot function as safety nets. Since PMTs measure assets and other ‘stable’ characteristics such as education levels, the PMT is a very static
mechanism. Even if it allowed people to apply for a social protection programme whenever they experienced a crisis, they are unlikely to b
e identified as ‘poor’ because their proxies may not have changed, despite a significant fall in income. As discussed earlier, they
would have to first divest themselves of many of their assets before they could be recognized as eligible. Yet, their loss of the assets would mean that their ability to recover
from the crisis would be compromised. So, although the World Bank refer to social
assistance programmes using PMTs as ‘social safety nets,’ they are nothing of the sort see Kidd, 2012
. A safety net should be available as soon as people experience a crisis: in contrast, a social assistance programme using a PMT will continue to exclude people who
have fallen into poverty until their assets have been depleted, which may take time. Social assistance using PMTs is, in effect, Poor Relief
– in other words, a programme for those who are already living in poverty
– and nothing more and is similar to the approach used by developed countries in the 19
th
Century. Advocates of PMTs have recently begun to claim that they can introduce
‘dynamic targeting
’ into PMTs.
11
By this they mean introducing on-demand applications to PMT rather than using one-off censuses. Yet, this will not solve the problem. It does not address
the in-built design errors nor the challenges of errors introduced during implementation; and, it will further increase the costs of implementing the PMT, with no guarantee that it
will be more accurate. Furthermore, as discussed earlier, even if households have been hit by a crisis and apply for a programme using a PMT, their assets are unlikely to have
changed and so they would continue to be excluded. And, it would mean selecting households using data collected at different times: for example, if mandatory
recertification of all households is only undertaken every five years, a targeting list using
supposedly ‘dynamic targeting’ would be a mix of data from households assessed at any time over a five-year period, meaning that like would not be compared with like.